Particularly, since the Go Match between AlphGo and Lee SeDol (the 18-time Go world champion) in March 2016, the popularity of Artificial Intelligent (AI) has swept over the world. The public’s interest and expectation in Deep Learning and Machine Learning have also grown. Let’s take a closer look at what is deep learning and machine learning, what is the relationship between deep learning and big data. Also, let’s find out the cases of deep learning.
What is Deep Learning?
Everyone may have experience of drawing a picture using a trend line in excel file. That is a very easy and simple way to use Regression Analysis in excel. As you might heard many times, regression analysis is one of mining methods for analyzing relationship between data and making a model formula. It has disadvantage that it is less accurate comparatively and much efforts are necessary for analyst to fine tune their model.
So, Artificial Neural Network (ANN) has appeared. That is a method that recreates the image of human brain. When data is entered in the type of black box, it automatically begins using a complex mathematical formula. For example, if we need to tell a dog from text, pictures, video clip and etc, ANN identifies the characteristics of dog and combines many factors to find the dog automatically.
Machine Learning means researching in algorithm to predict and classify based on property learned from Training Data. In other words, ANN is a field of machine learning. Many scholars have conducted researches to improve the accuracy of ANN. In particular, the accuracy of model formula has improved thanks to optimization theory and many Kernel functions.
Also, with the advent of big data technology, modeling can utilize much data improving their accuracy. Here, we call the convergence of big data with ANN as Deep Learning. So, deep learning is a kind of machine learning.
In more details, ANN needs to know whether an image in pictures is a dog or not in advance upon classifying a dog. Human needs to define the characteristics of dog in advance.
However, deep learning groups and classifies a dog automatically from many kinds of pictures with or without dog. In other words, computer itself trains on their own to find and classify a pattern of dog automatically without human efforts. That has been possible because of big data technology.
“Just for a second!”
What is different between Data Mining and Machine Learning?
Simply speaking, while data mining is focusing on finding the characteristics of current data, machine learning is focusing on learning now with the current data to predict the future. For example, when classifying a dog, data mining finds main characteristics of dog such as ears, mouth, color, on the other hands, machine learning predicts a dog in new pictures based on already-held data.
Voice and image recognition are mainly used in deep learning. They are using a lot of data with a relatively high accuracy, so many companies are providing relevant services by using deep learning tech. A good example in case is Facebook and Google.
Any Facebook user might understand it easily. When uploading a picture taken with your friends, the name of your friends is automatically tagged. That is an algorithm of Deep Face, recognizing and classifying the faces of friends using deep learning technology.
The accuracy of recognition is 97.25%, which is similar to that of human’s (97.53%). Deep learning is also used in advertising, which provides a customized advertisement for users after analyzing a product in pictures posted.
Google is using deep learning in photo tagging and voice recognition.
Google Translator, which was developed long time ago, has adopted deep learning, thus increasing their accuracy more and more. Google is also providing a service to make a photo album automatically by recognizing users’ pictures and sorting them out.
Baidu, called as Chinese version of Google, is using deep learning to strengthen their functions of voice recognition, image recognition and image searching. In Korea’s case, Naver has introduced deep learning tech in voice searching. Also, deep learning is getting widespread more and more in summarizing news and analyzing images.
Early this year, AlphaGo developed by Google has swept through the world with a growing popularity of Artificial Intelligence. AlphaGo is also a case using Deep Learning.
Also, self-driving cars are using deep learning. It has not yet been popular due to safety issue, but technology has significantly developed compared to the previous years. In the past, when a car passes through intersection, human had to make and enter all questions, for instance, are there people in the street? Are any car nearby? Is traffic signal turned green?
However, now we enter a large amount of video and photo data on a risky and normal situations when a car passes through intersection, then make computers learn the situations for themselves and to decide whether to pass or not automatically. The same goes to AlphaGo. All situation data on Go are entered in AlphaGo to make them decide on their own automatically. Difference is to calculate the chances of winning for all possible conditions and to choose the number with the highest possibility. (Use the method of Monte-Carlo Tree Search)
The technical development of ANN and big data dealing with massive amount of data have started to play a big role in our life. However, the success of deep leaning depends on whether to have a huge amount and various types of data or not. Until now, computers are not able to decide automatically on a situation which has not yet happened or has almost never happened. Due to this defect, Lee Se Dol could have a chance to win a couple of his matches with AlphaGo.
The same goes to an accident during self-driving. Even if a lot of video data on accident and dangerous situations are held, an accident that we cannot imagine might happen. To apply deep learning successfully, diverse data need to be collected and a way to respond to unknown situation need to be considered.
Even if enormous data is available, it needs to have much efforts and money to utilize them with deep-learning technology, to fine tune the data available to increase their accuracy up to the level where human can be satisfied with.
▶ The contents are protected by copyrights laws and the copyrights are owned by the creator and Samsung SDS.
▶ Re-use or reproduction as well as commercial use of the contents without prior consent is strictly prohibited.